Yeah, yeah. I said I would no longer waste blog space on fickle betas and unavailable alpha experiments… But lighting is the single most important aspect of any 3D film – no matter the art style, without highlights and shadows objects have no depth and cannot be visually “anchored” in a scene. Lighting also happens to be the framerate killer for game engines. Each scene light increases the scenes polygon count geometrically. Finding an artistic balance between the fewest possible live scene lights, and an approximation of ambient and reflected bounce lighting or global illumination, is the Holy Grail of realtime 3D…. Here’s the latest maybe…, called HXGI by the author of Hx Volumetric Lighting.
The typical solution is to bake static scene lights and shadows onto a second texture map, either in another program or internally in the game engine. Most game engines also employ reflection probes and a low-res dynamic shadow map to blend over the baked shadows. With the adoption of Lightmass in Unreal and Enlighten in Unity, small scenes can be entirely dynamic with propagated bounce light continuously rebaked on the fly, limited only by the resolution of their shadow maps and speed of the GPU. Very fine-detail shadows that fall below the resolution of the lightmap are handled with an onscreen image-effect ambient occlusion, and some models may have their own AO maps.
One problem however is that animated meshes can’t possibly rebake their maps every frame. A figure walking through the environment, including hair, clothing, and props, can’t update reflective GI maps quickly enough. Static scene objects can be pre-baked or updated every few seconds, but animated figures still require scene lights. A workaround is to use a Light Probes Proxy Volume: an array of point lights that take on a color approximating the bounce light. You can imagine what having dozens of extra scene lights approximating ambient bounce lighting will do to your frame rates. Such an elaborately lit scene would be a chore to create by hand, and cannot be generated procedurally with any strategic efficiency.
Voxel-Based Global Illumination
In old-fashioned ray-traced rendering, the camera traces a vector to each visible surface, and calculates the global illumination for each pixel by following the ray back to the light source(s). Game engines can’t possibly raytrace every pixel in realtime. What’s needed is a way to simplify the process with fewer rays, and bake the lighting values into an accessible “map” of volumetric elements, or voxels. If a rasterized imagemap is created from regularly-spaced pixels, a voxelized scene is represented by regularly-spaced cubes.
The automagic voxelization process subdivides a scene into diminishing cubes, discarding the empty spaces while dividing into smaller and smaller elements. The scale and subdivision of the voxels is adjustable, and not at all dependent on the polygon detail of the scene. Distant objects like mountain terrains can voxelize just as quickly as nearby, detailed models. Finally, each voxel creates a fast look-up reference map of the light in all directions using similarly reduced low resolution light “cones”. The engine generates only the voxels it can see through the camera, and data is interpolated from voxel to voxel, updating as needed. The result is voxel-based global illumination!
With physically based rendering (PBR), every surface material is reflective. The difference between a polished-mirror and matte skin is only in the brightness and blurriness of their reflections. With traditional raytracing realistic blurry reflections are computationally expensive, however voxel GI interpolates blurry reflections with a lower voxel resolution.
Lexie Dostal, creator of HXGI says, “One of the big issues with Enlighten is that dynamic objects are lit by a single interpolated light probe. this can make large dynamic objects look extremely out of place. The way my system works is each fragment (pixel) samples the GI data independently. this means large dynamic objects will be lit correctly.”